Assessment of Oxford Nanopore whole genome sequencing for large-scale genomic characterisation of Staphylococcus aureus

This study demonstrates that Oxford Nanopore whole-genome sequencing is a viable and often superior alternative to Illumina for large-scale genomic characterization of *Staphylococcus aureus*, offering improved detection of clinically relevant genes and variants despite minor lineage-specific error variations.

The Gist

Imagine you are trying to read a very long, complex instruction manual for a specific type of bacteria (Staphylococcus aureus) that causes blood infections. To understand how this bacteria works, what drugs it resists, and how dangerous it is, scientists need to read its entire "genome" (its genetic instruction manual).

For a long time, the only way to do this was like reading a book by tearing out every single page, shuffling them into tiny piles, and trying to reassemble the story based on the first few words of each page. This is Illumina sequencing. It's incredibly accurate for reading individual words, but because the pages are so small, it's hard to figure out the order of the chapters or find repeated paragraphs.

Recently, a new technology called Oxford Nanopore (ONT) came along. This is like having a super-fast scanner that can read the book in long, continuous streams, page after page, without tearing it apart. It's great for seeing the whole picture, but sometimes the scanner gets a little blurry or skips a letter here and there.

The Big Question:
The scientists in this paper asked: Can we trust this new "long-read" scanner (ONT) to handle a massive library of 836 different bacteria samples, or do we still need the old "short-read" method (Illumina) to get it right?

The Experiment: A Race Between Two Readers

The team took 836 bacteria samples from patients with bloodstream infections and ran them through both machines. They then compared the results to see which one gave a better story.

Here is what they found, using some simple analogies:

1. The "Blurry" vs. The "Broken" Book

• Illumina (The Short-Reader): It reads very clearly, but the pages are tiny. When the bacteria has a section with lots of repeated text (like a chorus in a song that repeats 10 times), the Illumina machine gets confused. It often misses the whole section or thinks there are fewer repeats than there actually are.
• ONT (The Long-Reader): It reads the whole song in one go. It handles the repeated sections perfectly. However, it occasionally misreads a letter (a "typo").
• The Result: The ONT machine was actually better at finding the full story, especially for complex parts of the genome. Even with a few typos, it gave a more complete picture of the bacteria's structure.

2. The "Special Ink" Problem
The scientists noticed that the ONT machine made more typos in certain specific bacteria lineages (families). They discovered this was likely because some bacteria have "special ink" (methylation) on their DNA that confuses the scanner. It's like trying to read a book written in invisible ink; the scanner gets stuck on those specific words. Interestingly, this happened mostly in one specific family of bacteria (ST25), suggesting different "families" of bacteria might need different reading strategies.

3. Finding the "Dangerous" Clues
The most important part of the study was looking for "danger signs"—genes that tell us if the bacteria is resistant to antibiotics or if it produces toxins.

• The Winner: The ONT machine found more of these dangerous genes than the Illumina machine.
• Why? Many of these dangerous genes are like "chapters" that are repeated over and over or are located on tiny, loose pages (plasmids) that can fall out of the book. The short-read machine (Illumina) often lost these pages or couldn't piece them together. The long-read machine (ONT) kept them all together.
• Specific Example: The ONT machine was much better at spotting the spa gene (a fingerprint used to identify the bacteria type) and various toxin genes that the other machine missed.

4. Do We Need to "Polish" the Book?
Usually, when you use the blurry scanner (ONT), you might want to use the clear scanner (Illumina) to fix the typos afterward. This is called "polishing."

• The Finding: The scientists found that polishing the ONT data didn't change the results very much. The ONT machine was already so good at finding the important genes that fixing the few typos didn't add much value.

The Bottom Line

This study is like a report card for two different ways of reading a book.

• The Verdict: The new Oxford Nanopore (ONT) technology is excellent for large-scale studies. It is faster, cheaper, and better at finding the "big picture" and the tricky, repeated parts of the bacterial genome that the old method misses.
• The Caveat: It's not perfect. It has a few typos, and it struggles a bit with specific "families" of bacteria that have special chemical markings.
• The Takeaway: For researchers trying to track outbreaks or understand how bacteria evolve, the long-read scanner is a powerful new tool. It might have a few smudges, but it gives you the whole story, whereas the old method often leaves out the most interesting chapters.

In short: If you want to know the full story of a bacteria, the long-read scanner is the better storyteller, even if it occasionally misspells a word.

Technical Summary

1. Problem Statement

Whole-genome sequencing (WGS) is critical for microbial diagnostics, surveillance, and research, particularly for large-scale comparative genomic studies (e.g., Genome-Wide Association Studies). However, researchers face a trade-off between sequencing technologies:

• Illumina (Short-read): Offers high base-calling accuracy (>99.99%) but struggles with repetitive regions, structural variants, and complete genome assembly due to short read lengths (150–300 bp). This often leads to fragmented assemblies and missed detection of genes located in complex regions.
• Oxford Nanopore Technology (ONT) (Long-read): Generates ultra-long reads (up to millions of base pairs), enabling complete genome assembly and resolution of repetitive regions. However, it has historically suffered from higher base substitution and indel error rates, raising concerns about its reliability for detecting single-nucleotide variants (SNVs) and specific gene calls without expensive and time-consuming hybrid polishing (combining with Illumina data).

The Gap: There was a lack of comprehensive, large-scale data comparing ONT and Illumina performance specifically for clinical Staphylococcus aureus isolates regarding genotyping, antimicrobial resistance (AMR), and virulence gene detection.

2. Methodology

The study performed a head-to-head comparison of ONT and Illumina sequencing on a large clinical cohort.

• Sample Collection: 836 unique S. aureus bloodstream infection (bacteraemia) isolates were retrieved from the Nord-Trøndelag Hospital Trust Sepsis Registry (collected 1996–2022).
• Sequencing:
  • ONT: Libraries prepared using the Rapid Barcoding kit (V14) and sequenced on a MinION Mk1B with R10.4.1 flow cells. Base calling used Dorado v0.7.2 (hac model). Reads were filtered (Filtlong) and assembled using Flye v2.9.4.
  • Illumina: Libraries prepared using Nextera XT and sequenced on HiSeq or NovaSeq (2x150 bp). Reads were trimmed (Trimmomatic) and assembled using Shovill (Nullarbor pipeline).
• Polishing: ONT assemblies were polished using Illumina data (Pypolca v0.3.1) to create a "gold standard" reference for error correction.
• Bioinformatic Analysis:
  • Genotyping: Multi-locus sequence typing (MLST) and spa typing were performed using mlst, spaTyper, and Prokka.
  • Gene Detection: AMR genes, virulence factors, and stress genes were identified using AMRFinderPlus and Abricate (with VFDB and PlasmidFinder databases).
  • Discrepancy Analysis: Raw reads from discrepant isolates were mapped back to reference genes to verify presence/absence and coverage. High-error regions were analyzed for sequence motifs.

3. Key Contributions

• Scale: This is one of the largest assessments of long-read sequencing for bacterial genomics to date, analyzing 836 clinical isolates.
• Practical Utility: Instead of a theoretical benchmark, the study focuses on practical outcomes relevant to clinical research: genotyping accuracy and the detection of clinically relevant AMR and virulence genes.
• Error Characterization: The study identifies that ONT errors are not random but are associated with specific sequence types (STs) and methylation patterns, challenging the assumption that polishing is always necessary for large-scale studies.
• Gene Detection Superiority: It demonstrates that ONT assemblies often outperform Illumina in detecting genes located in repetitive regions or low-GC content areas, even without hybrid polishing.

4. Key Results

A. Assembly Quality and Genotyping

• Assembly: ONT produced near-complete genomes (96.5% of isolates had a single circular chromosome >2.65 Mbp) with a median of 2 contigs, compared to Illumina's median of 103 contigs.
• MLST: Sequence typing (ST) concordance was high between technologies (88.5% for ONT vs. 88.8% for Illumina).
• spa Typing: ONT significantly outperformed Illumina. ONT successfully assigned spa types to 95.3% of isolates, whereas Illumina failed to type 23.0% of isolates because the spa gene (highly repetitive) was fragmented across multiple contigs.

B. Base Error Rates and Methylation

• Error Rates: After polishing, the median error rate was low (2.6 base errors and 3.6 indels per 1,000,000 bp).
• Lineage-Specific Errors: Error rates were not uniform. ST25 isolates showed significantly higher error rates (51.7 base errors/1M bp).
• Cause: Motif enrichment analysis revealed that high-error regions were associated with specific DNA motifs (e.g., DWGGWCCWH), likely linked to lineage-specific methylation patterns (e.g., M.Sau961) that confuse the ONT base caller.

C. Gene Detection (AMR and Virulence)

• Overall Concordance: 99.99% of genes detected in polished assemblies were found in raw ONT assemblies.
• ONT Superiority: For 42 genes/variants where technologies disagreed in ≥5 isolates, ONT had the higher detection rate in 39 cases (20.6%).
• Repetitive Regions: ONT detected significantly more copies of genes in repetitive regions (e.g., clfA, clfB, esaG, sdrC) compared to Illumina, which often collapsed these repeats.
• Low GC Content: Genes with low GC content (e.g., staphylococcal enterotoxins like seu, sen, seo) were frequently missed or under-represented in Illumina assemblies due to GC-bias in library preparation (Nextera XT). ONT detected these more reliably.
• Specific Discrepancies:
  • 23S rDNA C2220T: Detected in 15 isolates by ONT but missed by Illumina (likely due to low allele frequency in the read pool).
  • Plasmids: Some small plasmids (e.g., carrying ermC) were lost in ONT assemblies, leading to false negatives for those specific genes.

D. Impact of Polishing

Polishing ONT assemblies with Illumina data resulted in minor changes to gene calling and genotyping. This suggests that for large-scale population studies, hybrid polishing may be unnecessary if the goal is gene presence/absence or typing rather than single-nucleotide variant calling.

5. Significance and Conclusion

• Feasibility for Large Studies: The study validates ONT as a robust, standalone technology for large-scale bacterial genomic characterization. It eliminates the need for costly hybrid sequencing in many scenarios.
• Clinical Relevance: ONT provides superior resolution for clinically critical repetitive elements (like spa and adherence genes) and low-GC virulence factors, which are often missed by short-read technologies.
• Methodological Guidance: Researchers must be aware that ONT error rates can vary by bacterial lineage due to methylation. While polishing improves accuracy, the raw long-read assembly is often sufficient for genotyping and gene detection.
• Future Directions: The findings highlight the need to understand lineage-specific methylation effects and to optimize library prep for low-GC regions. The study supports the shift toward long-read sequencing for comprehensive bacterial surveillance and outbreak investigation.